COSMO version 1.0 (Release date: 4/11/06)

For further information on how to interpret this output, please access http://cosmoweb.berkeley.edu.

You can download these results in XML format here.

If you use this software in your research, please cite

O. Bembom, M.J. van der Laan (2006). Supervised detection of conserved motifs in DNA sequences with cosmo. UC Berkeley Division of Biostatistics Working Paper Series. Working Paper 209. http://www.bepress.com/ucbbiostat/paper209.

cosmo makes use of the donlp2() function by Peter Spellucci. The use of donlp2 must be acknowledged in any publication which contains results obtained with cosmo or parts of it. Citation of the author's name and netlib-source is suitable.

This information can also be useful in the event you wish to report a problem with the cosmo software

command: cosmo cosmo.seqs.13573 -minw 7 -maxw 12 -zoops -tcm -addfree -con cosmo.cons.13573 -xml -status

Order of background Markov model chosen by likelihood-based CV: 1

Kullback-Leibler divergences for candidate orders 0 to 6:

Estimated transition matrix for order 0

Estimated transition matrix for order 1

---

Explanation of cosmo results

---

The cosmo results consist of:

• Basic information about the version of cosmo, the date it was released, a link to your output in XML format, the date your job was run, the date this web page will be removed from the server, the time it took to process your job, and the server it was processed on.

• The reference to cite if you use cosmo in your research.

• A call summary detailing the parameters with which you ran cosmo.

• A description of the sequences you submitted (the "training set") showing the name and length of each sequence.

• A description of the constraints you submitted. The left column shows the original constraint definitions you supplied, and the right column shows the constraint definitions as they have been interpreted by cosmo. You may compare these two columns to ensure that the constraints you supplied were interpreted correctly by cosmo. Note that the interpreted constraints will contain an additional "empty" constraint set of the form

  @ Constraint set
  >IntervalSetup
  Length:variable
  

  if you did not supply any constraints or if you checked the option to add the unconstrained case to the given constraints.

• A description of the Markov model that was used to model the distribution of background nucleotides. In a Markov model of order k, the distribution of a nucleotide only depends on what nucleotides were observed in the previous k positions. By default this order is chosen by cosmo through likelihood-based cross-validation. In this case the output contains a table of Kullback-Leibler divergences for the various candidate orders. cosmo chooses the order with minimum Kullback-Leibler divergence. Furthermore, the output shows the estimated transition matrices up to at most order 2. Each entry shows the estimated probability that a given prefix (shown at the left) is followed by a particular nucleotide (shown at the top of the column).

• A summary of the candidate models that cosmo considered. cosmo can choose data-adaptively among a large number of models for your data that can differ in the constraints on the position weight matrix, the distribution of sites, the motif width, and the number of motif occurrences. cosmo chooses these parameters in a sequential fashion such that for each given constraint set, distribution, and motif width, it will first identify the optimal number of motif occurrences. Then, for each given constraint set and distribution, cosmo will find the optimal motif width (with a corresponding number of motif occurrences that has already been selected). In the next step, an optimal distribution is chosen for each given constraint set. Finally, the best constraint set is selected. The table of candidate models summarizes the values of the relevant selection criteria for choosing the motif width ("wCrit"), distribution of motif occurrences ("distCrit"), constraint set ("conCrit"). Given the sequential fashion in which parameters are selected, not all combinations of parameter values have been evaluated so that some elements of the table will be left blank. cosmo selects parameters by minimizing the relevant criterion across candidate values.

• A summary of the selected model that shows the selected constraint set, distribution, motif width, and number of sites along with the corresponding criterion and optimal criterion value for parameters that have been chosen data-adaptively.

• The estimated position weight matrix. Its entries show the estimated probabilites with which each nucleotide occurs in each of the positons of the discoverd motif.

• The sequence logo of the discovered motif. The height of each column is gives the information content at the corresponding position in the motif. The heights of the individual letters are proportional to the estimated nucleotide frequencies. Sequence logos are generated by the WebLogo program of Crooks et al.

• An alignment of the discovered sites. The output includes the E-value given alignment and gives the following information for each discovered site:
  • The name of the sequence on which the site was discovered.
  • The strand of this sequence on which the site was discovered. A + indicates the given (positive) strand, and a - indicates the reverse complement strand.
  • The position on the sequence in which the motif starts
  • The posterior probability, given the estimated position weight matrix, that this site represents an occurrence of the motif.
  • The actual site. The output includes a number of surrounding nucleotides (lower-case) in addition to the site itself (upper-case and in color).

• A plot of posterior probabilites along each of the sequences. Bars are color-coded to represent the nucleotide at a given position: Green for A, blue for C, orange for G, and red for T. The bar at a given position extends upward if the posterior probability is greater for a motif occurrence on the forward strand and downward if the posterior probability is greater for a motif occurrence on the reverse strand.